Beyond Tags: Content Discovery with Tripadvisor's Hybrid Graph Search
summary
Tripadvisor builds a hybrid retriever inside Neo4j combining structural tag lookups with destination-scoped vector search, using per-destination vector indexes, HNSW approximate nearest neighbors and MPNET/NGT embeddings to keep latency predictable. The pipeline rescored candidates with a cross-encoder ranker and an LLM judge, updates the graph each cycle, and scales across tens of thousands of geos and tags, hundreds of millions of locations and over a billion reviews, ultimately restored thousands of missing tag POI connections and introduced over 100 new tag concepts.
AI-written summary · about
chapters (7)
- 0:00Introduction and team
- 1:08About TripAdvisor and tags
- 4:03Hybrid graph search approach
- 8:44Vector indexes and embeddings
- 13:00Combining structured and semantic search
- 15:51Results and self-enriching graph
- 18:33Q&A session
related talks
17mEnhancing Chat-Based Customer Recommendations With GraphRAG and Hybrid Search
15mGraph-Driven Knowledge Retrieval: Neo4j for Healthcare Building Codes
01h 17mGraph Database and Retrieval Augmented Generation
42mIntegrating Graph Context With OpenSearch for Enhanced RAG Application Performance
01h 00mGraph Mastery to AI
27mEventKernel: Multi-Hop Reasoning and GraphRAG for AI-Powered Event Intelligence
25mEnhancing Job Matching With Knowledge Graphs and RAG
13mOptimizing Real-Time Job Matching: Integrating Neo4j With RAG